Pauli Virtanen

I think that's probably not so crucial, it would probably also be fine to shell out to a subprocess for the env creation.

The workaround to the nested virtualenv/venv case apparently requires finding the "real" Python executable: from https://github.com/berdario/pew/blob/c25f19e9bb0bc94f696407378fc104706b5dd3b5/pew/_venv.py#L68-L104 ``` def find_real_python(): # We need to run venv on the Python we want...

There is a suggested workaround to Pipewire for SOF cards for xrun recovery failures: https://gitlab.freedesktop.org/pipewire/pipewire/-/merge_requests/2548 Is the problem related to this one? cc @ujfalusi

Headroom for drop/prepare/start is not quite right in PW for xrun recovery, its done form RT thread, and if the RT thread blocks it blocks the whole audio graph also...

@ujfalusi see alsa-pcm.c, spa_alsa_start/open that open device and do cold start run from main thread (blocking long is OK), alsa_recover from RT thread (blocking long not OK). Data aside from...